We introduce Rapid Simulation Based Inference (RSBI), a diffusion-based variational approach to likelihood-free Bayesian inference that achieves high sampling efficiency under large prior-to-posterior volumes. RSBI builds on advances in Schr\"odinger Bridge (SB) diffusion sampling to handle multi-modal posteriors, with an initial likelihood supplied by a distance measure via a differentiable simulator. The key innovation of this initial surrogate likelihood is that it gives a highly efficient proposal distribution, replacing sequential rounds of other methods with an efficient one-shot proposal. Subsequent refinement then transports the proposal to the true posterior as specified by the simulator and the original prior. The SB sampler is warm-started from the proposal, and its energy is corrected by a neural likelihood ratio that is learned only where the proposal places mass. As a variational method, RSBI does not suffer from prior leakage commonly observed in standard sequential posterior estimation methods. To improve mode coverage we optionally utilize the well-tempered meta-dynamics framework, which also encourages exploration of the prior volume. We observe strong performance on standard SBI benchmarks, particularly when the prior volume is increased relative to the ground truth posterior, achieving a performance degradation of only $\sim5\%$ on Two Moons under a highly constrained simulation budget. Additionally, we evaluate RSBI's performance for gravitational wave ring-down posterior estimation, in a real-world physics benchmark inspired by black hole spectroscopy.
Hierarchical multi-label text classification (HMTC) requires models to capture both structural dependencies among labels and the fine-grained semantics of labels at different hierarchy levels. Existing methods often model hierarchy through prediction constraints or representations of individual labels and local relations. Structural and semantic supervision, however, are usually formulated separately. We instead use complete root-to-node label paths as a shared unit for both. Our framework introduces two auxiliary training objectives while retaining a standard multi-label classifier at inference. PathSimNCE aligns document representations with hierarchical paths using hierarchy-aware contrastive targets. These targets provide graded supervision for structurally related paths while preserving explicit supervision for the ground-truth paths. The complementary semantic objective uses LLM-generated, document-conditioned descriptions of the same positive paths. It thereby provides context-specific label semantics beyond static label representations. Experiments on three benchmarks show consistent improvements: structural alignment is particularly beneficial in deeper hierarchies, while semantic alignment provides additional gains for fine-grained labels.
Given $n$ i.i.d. random matrices $A_i \in \mathbb{R}^{d \times d}$ with common expectation $\Sigma$, the goal of \emph{Differentially Private Stochastic PCA} is to identify a $k$-dimensional subspace capturing the leading variance directions of $\Sigma$, while preserving differential privacy~(DP) for each individual sample $A_i$. \citet{dungler2025iterative} introduced \kdppca, the first algorithm to simultaneously (1) achieve sample complexity $n=\widetilde O(d)$ for sub-Gaussian data, (2) adapt its privacy noise to the intrinsic randomness of the data, and (3) extend seamlessly to any target dimension $k\le d$. However, its sample complexity has a suboptimal dependence on $k$. We propose the first algorithm that achieves optimal sample complexity in both $d$ and $k$, while retaining properties (2) and (3) of \kdppca. Additionally, prior utility guarantees require a minimum sample size with exponential dependence on the eigengap. Our method removes this dependence. It also improves the dependence on spectral parameters, matching known lower bounds for the spiked covariance model. Unlike deflation-based approaches like \kdppca, our algorithm updates the full $d\times k$ subspace jointly rather than one eigenvector at a time. This non-deflation structure simplifies the algorithm, reduces the number of hyper-parameters, and improves computational efficiency, particularly when using block linear algebra libraries.
Coordination between large language-model agents has become a central question for AI safety and software engineering, as such agents are deployed in groups that share infrastructure. Existing research tells the agents that partners exist or asks them to coordinate. Whether coordination arises through shared state alone, with nothing disclosing that anyone else is present, is the key question for deployments that share caches and memory between agents, yet it has rarely been examined, and never with independent replication; this study does both. Four agents each received a short arithmetic task and a key-value store presented to each as its own scratch space; two held complementary halves of one computation, and no prompt mentioned other agents, sharing or collaboration. For reliability, the environment was rebuilt independently from a written protocol, and both implementations ran under hashed locks on design, data and analysis. The results bear novelty value in three respects. Joint success occurred only through the shared store. Whether a model coordinated depended not on its vendor tier but on whether it wrote its intermediate result and looked at the store again, and one ordinary sentence, "check your workspace again before submitting", switched that second look on and off. The findings held under replication. For practice, state shared between agents is a communication channel and should be separated, namespaced and logged. For research, designs should be locked before the data, and the parse rule, endpoint and parameters reported. Future work should test open-weight models and separate using another agent's output from intending to communicate.
While modern video generative models achieve remarkable visual fidelity, they remain noticeably brittle when generating physically plausible dynamics and complex interactions. Targeted post-training is crucial, yet supervised fine-tuning suffers from offline state mismatch, and reinforcement learning is hindered by sparse, ill-defined physical reward functions. On-Policy (Self-)Distillation (OPSD) reconciles on-policy rollouts with dense corrective feedback, shifting the core challenge to designing effective privileged context. However, conventional modalities fall short: text prompts are inherently declarative, whereas real reference videos heavily entangle physical laws with visual appearance without parametric control. To this end, we propose PhysOPSD, introducing executable code as privileged physical context for video generation. Unlike static visual observations, executable code instantiates an editable physical world with intrinsic abstraction and controllability. Powered by PhysOPSD-Engine, we translate real-world physical scenarios into verifiable simulation programs. PhysOPSD then incorporates (i) dual-level physical privilege to bridge the information gap via an intermediate bottleneck, alongside (ii) controlled physical variations under shared rollout noise to expose causal dynamic shifts. Extensive experiments demonstrate that PhysOPSD substantially improves physical plausibility across diverse backbones, outperforming SFT and RL baselines with only $3$K samples, while preserving general video generation quality.
Robots are increasingly expected to execute natural-language manipulation tasks in real-world environments. This is particularly relevant in household assistance and logistics, where clutter requires reasoning about both task objectives and execution feasibility. This challenge becomes even more pronounced for bimanual humanoid platforms, where coordinating two manipulators introduces additional interaction and kinematic constraints. We present a zero-shot vision-language framework for execution-aware bimanual manipulation that jointly reasons about task objectives and physical feasibility. Planning is grounded in a structured scene representation that combines semantic object relations with execution constraints, enabling the selection of obstacle relocations and execution choices while preserving task semantics. Compatible actions are organised into parallel bimanual stages with explicit pre- and post-conditions. A hierarchical validator evaluates these conditions under perceptual uncertainty and uses stage-level execution history to support local recovery or targeted replanning. Experiments on the CENTAURO hybrid wheeled-legged humanoid robot across nine language-conditioned manipulation scenarios show higher task success than a non-execution-aware baseline. In dedicated failure scenarios, hierarchical validation recovers from execution failures with fewer model queries than always replanning. These results demonstrate the benefits of combining execution reasoning directly into task planning. Videos and additional material are available on the Project Webpage: https://advrhumanoids.github.io/bridging-planning-execution/.
Despite growing attention to social bias in large language models (LLMs), existing bias benchmarks provide limited coverage of culturally grounded evaluation in Bengali, particularly within Bangladesh's sociocultural context. To address this gap, we develop a bilingual extension of BanglaBBQ comprising 61,312 question-answering instances across 12 social categories in English and Bengali. The benchmark incorporates balanced answer permutations, contrasting question framings, and both ambiguous and disambiguated contexts to systematically examine model performance and stereotype-aligned bias across languages. We validate the dataset through structural checks, 220 statistical tests, and expert annotation of 150 instances by four annotators, achieving strong inter-annotator agreement (Fleiss' \k{appa}=0.8349). We evaluate eight LLMs across 490,496 instances and identify five key findings. First, disambiguated contexts yield substantially greater accuracy improvements in Bengali (+36.59 percentage points) than in English (+17.71). Second, all evaluated models exhibit lower accuracy in Bengali, although Mistral-Nemo shows the smallest cross-lingual accuracy gap (5.12 percentage points). Third, stereotype-aligned bias exhibits language and model dependent variation, increasing in Bengali for Gemma 2 9B but decreasing for Mistral-Nemo. Fourth, higher accuracy does not necessarily correspond to lower stereotype-aligned bias. Finally, bias varies considerably across social categories, with Qwen 2.5 7B exhibiting stronger stereotype alignment in Bengali for Politics and Physical Appearance. These findings demonstrate that cross-lingual performance and social bias are distinct dimensions of LLM behavior and underscore the importance of culturally grounded, language-sensitive benchmarks for evaluating fairness beyond English-centric settings.
An edge video analytics system must estimate the quality of local detections since there are no ground-truth labels available during runtime. Detectors usually provide confidence scores for their results, but these scores can not reveal every missed object or misplaced bounding box. So a quality estimator is necessary to provide a decision signal indicating which frame has potential detection failures that need to be escalated to a more powerful model in the cloud. Neural network based estimators add workload to the GPU already running detection, while CPU capacity may remain available. We present Predictive Temporal Consistency IoU (PTC-IoU), which moves this workload to CPU by compacting detection boxes histories to support deterministic result scoring. Comparing to neural counterparts, experiments on representative edge hardware, Nvidia Jetson Orin Nano, show that our method reduced per-frame estimator latency from 10.3 ms for neural counterparts to 2.2 ms (4.7x faster), increased pipelined throughput by 13.6%, and cuts estimator energy consumption by up to 95%, while maintaining competitive final detection quality verified on multiple datasets.
As LLM agents become more autonomous, they are increasingly expected to run as always-on assistants on a user's behalf, identifying work to be done and completing it without being asked. Yet existing benchmarks usually give the agent its task, whereas in knowledge work the requests sit in the user's workspace and the agent must first recognise which obligations exist. We study proactive task discovery and execution, in which an agent must infer a user's pending tasks from their workspace and complete them in the systems where the work lives. This setting poses two challenges. For the agent, the tasks are buried in a large volume of mail, chat, documents and records. For evaluation, a benchmark needs a simulated organisation whose people, events and documents are mutually consistent, so that tasks arise realistically and their correct outcomes are known. We introduce AutoBuddy-Bench, a simulator that addresses these challenges. It models an organisation's employees, projects, business systems and event history as a temporal graph, renders it into realistic workspace artefacts, derives ground-truth tasks from the events that create obligations, and computes each task's outcome from the world state; execution is judged by the agent's changes to the workspace. To evaluate current LLMs, we build AutoBuddy, a reference agent with interchangeable backbone LLMs: it identifies tasks from the workspace, grounding each in its source records, and executes each in a separate session with write access. Varying only the backbone, we find a systematic shift in where tasks are lost: weaker models fail mainly at discovery, intermediate models identify tasks but often stop short of carrying them out, and the strongest models fail mainly during execution, most often on business-system operations. Discovery precision is only weakly related to execution quality, suggesting the two are distinct capabilities.
In addition to the effects of explicit input prompts and instructions, large language model outputs are influenced by underlying behavioral preferences similar to human personality tendencies. These tendencies can affect how models interpret situations and make decisions. However, their influence is difficult to disentangle from the effects of model capabilities and instructions. Therefore, we introduce SPACE, a benchmark framework combining a controlled evaluation protocol with complementary datasets for personality verification, capability assessment, and hierarchical situational tasks. Controlled comparisons across guidance conditions help distinguish personality influence from measured capability changes and immediate instruction effects. Experimental results demonstrate that although prompting can shift generation style, models' internal personality tendencies are persistent and not readily overridden by opposing instructions. Furthermore, these underlying tendencies have an observable influence on LLM task performance, highlighting the need for greater attention to these characteristics and potential applications. Training code and SPACE benchmark specifications are available at https://github.com/Niklauseik/SPACE-training.
We study the fingerprinting of agentic cyberattacks. Our objective is to reliably infer information about the attacker's configuration, such as LLM version and harness, from telemetry. To achieve this, we introduce a technique we call shibboleth traps: modular honeypot services designed to elicit discriminative behaviors from agentic attackers. The traps make informative interactions a condition for obtaining resources the agent seeks, eliciting model declarations and characteristic writing without requiring a successful prompt injection. To achieve harness fingerprinting, we introduce general mechanisms to indirectly probe peculiar execution behaviors of harnesses. These probes reveal how a harness masks secrets in tool output, enforces command time limits, and schedules tool calls, using only the effects visible in subsequent requests. We show that combined declarative and behavioral inference distinguishes among 15 candidate models with an average 90% top-1 accuracy, with behavioral classifiers evaluated on held-out harnesses or task families. In a separate experiment across four harnesses, fixed probe rules identify the correct harness with 92% accuracy. When instructed to conceal their identity, agents still leave useful behavioral evidence even as their declarations become uninformative.
Artificial intelligence models for clinical prediction are proliferating across neurosurgery, but their aggregate discriminative performance, distribution along the care pathway and methodological quality are unclear. We systematically reviewed and meta-analysed these studies in neurosurgery and its subspecialities along with their risk of bias appraisal. We performed a PRISMA search for peer-reviewed original studies using AI to predict a clinical outcome in neurosurgical populations. Metrics were pooled using random effects modelling. Each study was classified as diagnostic, prognostic or forecasting and was assigned an evidence level. Clinical prediction task sub-stratification analysis was performed. Risk-of-bias was assessed using PROBAST. 1063 studies were included. Across 451 studies reporting AUC, pooled discrimination was 0.86, ranging from 0.84 (spine) to 0.87 (neuro-oncology), with extreme heterogeneity. Performance varied by clinical pathway stage, highest for diagnosis, grading and molecular tasks and lowest for post-operative tasks. Only 51 studies were using prospective data with external validation, whereas 619 rested on internal validation alone or no validation. 89 percent studies were at high risk-of-bias, driven by the analysis domain. AI models report high performance across all the neurosurgical subspecialities, concordant across AUC, accuracy, sensitivity and specificity and robust to risk-of-bias stratification. Performance is highest for imaging-anchored diagnostic tasks and lowest at decision-facing post-operative stage. However, the literature is dominated by optimism-prone, internally-validated and poorly calibrated models with extreme heterogeneity. External validation, calibration analysis and reporting guidelines adherence remain prerequisites for clinical adoption.
Autoregressive VLA planning usually predicts continuous trajectories as the final output, but driving instructions often describe actions rather than ego states, which weakens direct conditioning on the final continuous trajectory. We address this gap with TrajCond, which introduces a discrete trajectory-geometric condition for VLA planning. Specifically, we represent TrajCond as a 1D-Brohan differential action trajectory and learn it with a two-stage training framework before final continuous trajectory prediction. With the simultaneous autoregressive generation of discrete and numeric trajectories, we name our method AutoVLA-D (Double). Experiments on NAVSIM navtest, nuScenes, and NeuroNCAP show that discrete trajectories are weaker as direct outputs but more effective as TrajCond, consistently improving final continuous trajectory prediction across closed-loop and open-loop driving benchmarks. Code is available at https://github.com/ShivaTeam/AutoVLA-D.
Micro-scale air quality models require hourly traffic data on every road segment, whereas permanent counters cover only a small part of the network. A graph neural network (GNN) was developed to reconstruct hourly traffic flows on the 32,163 road segments of Paris intra-muros from the 1,573 segments equipped with a counter (4.9%). The model adapts the Graph Neural Network for Urban Interpolation (GNNUI) to a real, sparse counting network. Flows are decoded on the edges of the junction graph (one edge per segment, both directions summed), a traffic climatology serves as an offset in a zero-inflated negative-binomial likelihood, and training proceeds by masking counters. Eight-fold spatial cross-validation, with counters sharing a published series merged before splitting, distinguishes interpolation at a withheld counter in an instrumented district (R${}^2$ = 0.89, error of a quarter of the mean flow) from extrapolation to a district without counters (R${}^2$ = 0.74, a third of the mean flow). These scores hold on main axes where counters are located, while relative errors are about 50% on the few counted local streets. Ablations show that the climatology offset tends to help extrapolation. Road attributes, time cycles and the live state of the counting network matter most, while weather inputs have no significant effect. The masking rate is the only inherited hyperparameter with a clear effect. When evaluated on January--August 2026 (outside the training period), the frozen model stayed around the error band of previous years.
Shadow IT, the use of technology that the IT organisation has not approved, is a long-standing governance problem. We argue that autonomous AI agents can make it grow much faster: agents make building tools nearly free, act with their user's privileges, and leave behind scripts, automations and MCP servers that are themselves unsanctioned. A simple compounding model shows that, under its assumptions, shadow IT grows exponentially once agentic artifacts reproduce faster than the organisation retires them; retiring them starts with detecting them. We then ask whether unsanctioned agents can be detected on Windows endpoints. We propose a taxonomy of nine classes of agents and six layers of endpoint footprint, and argue that rules based only on the structure of processes and connections cannot separate an agent from its behavioural twin, the same program talking to a non-AI service. We present Agent Detector, an open-source Windows detector that anchors behaviour to the identity of the remote service and correlates it with configuration, installation and execution history, using a catalog of 39 tools and 184 LLM endpoint entries; it informs users through a notice they must acknowledge instead of blocking. On a controlled bench of 12 synthetic agents, 3 AI components and 9 benign workloads, repeated 5 times on one workstation, two behavioural rules flagged every behavioural twin, while Agent Detector flagged none and detected every agent in the majority of runs. Its limits are temporal: a connection to a provider behind a shared content-delivery network is attributable only while the DNS cache holds the provider's name, one to five minutes, and an intermittent agent was seen by about one scan in three. A scan costs about 154 ms. The bench is too small to rank detectors; it maps what each kind of evidence can see: unsanctioned agents can be partly detected, and not by behaviour alone.
We introduce AudioCodex, a fully causal autoencoder for live generative audio applications that compresses stereo audio into compact continuous latent sequences at ~10 Hz. It combines a Mean Flow magnitude autoencoder with latent hypersphere projection, bounded-noise augmentation, representation alignment, and adversarial phase recovery to learn representations that support both efficient downstream generation and high-quality reconstruction. Through extensive ablations, we show that AudioCodex outperforms the state-of-the-art in terms of generation-reconstruction trade-off, whilst remaining strictly causal in both encoding and decoding. This work paves the way towards truly real-time interactive and streaming generative audio applications.
Correlation Matrix Memories (CMMs) store associations as sums of rank-one outer products, providing closed-form addressability, exact deletion, and linear attribution. While classical analyses assume random, orthogonal keys, semantic encoders produce correlated representations. We analyse CMMs under correlated keys using frame-theoretic coherence: (i) capacity bounds yielding an encoder-independent limit $n \le d(1+\varepsilon^2)$ for mean squared crosstalk $\varepsilon^2$; (ii) an exact characterisation of the collateral effect of deleting an association, guaranteeing that deletion never invalidates certified associations; and (iii) a geometric transparency bound demonstrating why attribution requires candidate sets of centred keys. On this basis, we introduce Topology-Aware Neuro-Symbolic Associative Retrieval (TANSAR). TANSAR traverses a directed acyclic association graph, prunes invalid paths via sound ground Horn constraints, and injects a representation derived strictly from the verified path into the residual stream of a frozen language model, guarded by a Clopper--Pearson selective-risk gate. Controlled experiments confirm all theoretical bounds. In multi-hop knowledge editing, Horn pruning reduces answer contradictions from 25.1% to 0.5%, whereas Hopfield retrieval incurs 24% contradictions when stale associations remain stored (and 0% if stale associations are removed). An end-to-end edit transaction requires 13.6 ms on a standard CPU and 2.37 ms on GPU. The symbolic component matches gold labels across all 600 ProofWriter theories and achieves 94.0% agreement on FOLIO, demonstrating sound and efficient multi-hop associative retrieval with exact unlearning.
Multimodal retrieval increasingly supports agentic systems, where user requests can involve cross-level relations, combine multiple constraints, and evolve across dialogue turns. In these settings, a highly similar result may still violate the user's requirements, while retrieved content may directly inform downstream actions without human screening. We introduce TRIAD, a multimodal retrieval benchmark for evaluating how well retrieval systems satisfy these complex, evolving requests. TRIAD comprises three tracks: abstraction, testing bidirectional retrieval between visual instances and concepts at different hierarchical levels; constraints, testing the joint satisfaction of positive and negative conditions; and state, testing the integration of requirements across dialogue turns and the updating of earlier conditions when users revise their requests. The benchmark contains six subsets, each with 1,000 queries and a fixed pool of 1,000 candidates per query, evaluated using a common multi-positive mean average precision (mAP) protocol. Simpler reference tasks and hard negatives grouped by the requirements they violate help characterize specific retrieval difficulties. We evaluate 20 configurations of 17 systems and find that no system leads across all six subsets. Performance declines with hierarchical distance, and negative-only queries generally underperform positive-only queries with the same number of constraints. Across paired tasks expressing the same final intent, multi-turn retrieval yields lower mAP in 39 of 40 configuration-domain comparisons. The benefits of explicit reasoning also vary across models and tasks. These findings highlight the gap between retrieving similar content and satisfying a complete request.
A key challenge in action-conditioned world modeling is to construct macro-states from high-dimensional observations that support recursive prediction under actions, while keeping their predicted evolution consistent with visual changes in generated observations. We propose QCC-WM, a visual world model based on Quotient-based Causal Construction. QCC-WM uses action responses to guide macro-variable construction for recursive action-conditioned prediction. A shared reference quotient measurement constrains the consistency between predicted states and observed changes, allowing macro-level predictions to provide structured constraints for future visual generation. Using translational and rotational actions and their predicted responses, the model aggregates local visual features into macro-nodes with spatial correspondence through causal abstraction, and recursively predicts node content and motion under a shared reference measurement. Macro-level predictions guide future observation generation, while predicted motion transports historical appearance information. Node-level errors between generated observations and macro-level targets are then mapped back to the corresponding spatial regions for correction, improving consistency between visual generation and macro-level predictions. On the US-VLA dataset, QCC-WM achieves an ROI-MSE of 0.01861 for four-step prediction. Compared with the best baseline at each prediction horizon, it reduces LPIPS by 48.59%, 41.68%, and 30.23% for four-step prediction and eight- and sixteen-step recursive extrapolation, respectively. Ablation studies and mechanism analyses show that motion-guided reference transport mainly improves pixel reconstruction, while node-level error feedback further improves perceptual quality and consistency between generated observations and macro-level predictions.
Achieving generalizable and well-aligned multimodal representations remains a core challenge in robot learning. While recent approaches have attempted to align modalities by modeling conditional or higher-order statistical dependencies, they do not explicitly consider the structural coherence across modalities. In this work, we propose a novel multimodal alignment method that augments existing contrastive losses with a geometry-aware Gromov--Wasserstein (GW) distance-based regularization. To this end, we encode intra-modal geometry for each modality and reduce cross-modal structural discrepancies through a dynamically learned GW barycenter, thereby encouraging structural consistency in addition to instance-level contrastive alignment. We apply this optimal-transport-based alignment strategy to robot perception tasks involving underexplored modalities such as force and tactile signals, where modality data often exhibit varying sample densities. Extensive experiments on five robotic benchmarks show that our method improves structural alignment and downstream performance over state-of-the-art multimodal contrastive baselines across a wide range of downstream tasks, including robot and environment state prediction. Moreover, our GW-based regularization can be seamlessly integrated into most InfoNCE-like objectives.